kube-scheduler doesn't "just pick a node" — it runs 14 plugins per node

Hinzugefügt: 2026-05-05

The Kubernetes kube-scheduler does not simply pick a node for pod placement; instead, it runs 14 in-tree filter plugins in parallel for every pod and every node, including NodeUnschedulable, NodeName, TaintToleration, NodeAffinity, NodePorts, NodeResourcesFit, VolumeRestrictions, NodeVolumeLimits, VolumeBinding, VolumeZone, PodTopologySpread, InterPodAffinity, DynamicResources, and NodeDeclaredFeatures, with each plugin returning a binary verdict that determines whether the node is schedulable or not.